ProtBed: a streamlined web platform for exploratory statistics and predictive modeling in quantitative mass spectrometry
Year: 2026
Type: preprint
Abstract: Mass spectrometry proteomics and metabolomics generate high-dimensional quantitative matrices requiring rigorous statistical and machine learning evaluation, yet existing analytical workflows typically depend on programmatic scripting or remote cloud servers that introduce datagovernance constraints. ProtBed is a client-side web application that executes post-extraction data exploration directly in the browser runtime. The platform integrates deterministic preprocessing, pairwise and omnibus hypothesis testing, six-model cross-validated machine learning, and publication-ready visualization while preserving complete local custody of quantitative abundance matrices and clinical metadata. Empirical validation was conducted across four diverse cohorts totaling 250 clinical specimens. On an untargeted GC-MS urine cohort (𝑛 = 40; 356 metabolites), stratified five-fold cross-validated classifiers discriminated diabetic kidney disease with AUC = 0.90 (89% sensitivity, 82% specificity). In a deep-coverage post-mortem substantia nigra tissue cohort (𝑛 = 30; 9,606 quantified proteins), full-matrix modeling achieved out-of-fold AUC = 0.99 using logistic regression (AUC = 0.96 for linear support vector machines) and prioritized neuroinflammatory mediators including SAA1, CRP, and GFAP. ProtBed further resolved multiclass diagnostic arms in plasma extracellular-vesicle proteomics (𝑛 = 124) and characterized dysregulated lipid classes in Alzheimer's disease plasma (𝑛 = 56). ProtBed is freely accessible at https://protbed.com/ .
Source: ChemRxiv
Author: Anas Bedraoui
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7222993716